You are the context-warm-sessions skill.
When to use
- Resuming long-running engagement after days off
- Cross-session memory recovery (operator forgot what was decided last week)
- Verifying assumptions against prior session decisions
When NOT to use
- Single-session work (overkill)
- Just want most-recent state —
/li:resumealready reads 00-state.md
Workflow
Step 1 — Find context saves
source_root="${LINTEL_SOURCE_ROOT:-${LINTEL_REPO_ROOT:-$(git rev-parse --show-toplevel)}}"
source "$source_root/bin/_context.sh"
branch=$(_context_branch)
N="${1:-3}" # default last 3
case "$N" in [1-5]) ;; *) echo 'Choose 1–5 sessions.' >&2; exit 1 ;; esac
# Shared legacy checkpoints must belong to the selected repository. Filter
# ownership before choosing the newest N, just as context-restore does.
candidates=$(context_list "$branch" | sed -n "1,${N}p")
Step 2 — Estimate tokens
Per file, estimate. Aggregate.
Step 3 — Confirm load
CONTEXT WARM — sessions for branch <branch>
Found <N> recent saves:
- 2026-05-27 22:00 (4.5k tokens) — last activity before today
- 2026-05-25 14:30 (6.2k) — DEFINE phase commit
- 2026-05-23 09:15 (3.8k) — research-dive on ExpressRoute
Total: ~14.5k tokens
Load all? (Y / select subset / cancel)
Step 4 — Delegate to context-warm
/li:context-warm <selected-files>
Step 5 — 00-state.md append
event: context_warm_sessions
branch: <branch>
sessions_loaded: <N>
tokens_added: <approx>
Integration
Reads .claude/runtime/sessions/<branch>/. Delegates to /li:context-warm.
Anti-patterns
- Loading >5 sessions — diminishing returns, just causes context bloat
- Cross-branch session load — usually wrong; sessions are branch-scoped